"I could carve out of a banana a Judge with more backbone than that!"-A disappointed Theodore Roosevelt, after hearing of Justice Holmes's vote in Northern Securities Co. v. United States (1)
Can the president confidently predict the ideology of a supreme Court appointee? Voters seem to think so. In 2008, polls found that the majority of voters considered appointments to the Court an "important factor" in determining their votes. (2) More than one in six voters considered it the most important factor, ahead of even war and the economy. (3) The desire to control the nomination process is not a recent phenomenon. More than two-thirds of registered voters listed supreme Court appointments as an important factor in deciding their vote in the 2000 presidential election, and the supreme Court has been a plank on presidential platforms since 1896. (4)
Voters' desire to influence the Court stands in tension with the notion that the Court is a counter-majoritarian institution. This view of the Court has fueled the legal academy's "obsessive" discourse (5) on the counter-majoritarian difficulty. (6) Still, political scientists and a number of legal scholars have questioned whether the Court is, in fact, counter-majoritarian. (7) Though Justices themselves are not elected, elected officials appoint them. Thus, at the time of appointment, Justices should reflect dominant political views. (8) Most arguments for a majoritarian Court depend on the president's and Senate's ability to appoint Justices whose decisions reflect their views. (9)
The president's ability to appoint ideologically compatible Justices is a critical issue for both voters and Supreme Court scholars. Is there any reason to think presidents actually possess this power? If they do, one would expect Justices appointed by the same president, and perhaps by presidents of the same party, to vote together at a higher rate. Likewise, if appointments bring the Court in line with majoritarian views, then at least in times of unified government one would expect Justices to align with appointees of the same president or party more often than not. (10)
But the results of presidents' Supreme Court appointments are mixed. Consider the different experiences of George H.W. Bush and his son. Both of George W. Bush's appointees, John Roberts and Samuel Alito, have similar voting records, which are thought to align with executive preferences. The first Bush Administration did not fare as well. While Justice Clarence Thomas votes with Republican appointees at a high rate, David souter voted with Democratic appointees at just as high a rate. (11) It is no surprise that appointees sometimes will deviate from executive preferences, but how often are such disappointments likely to occur? Was Souter's appointment a product of divided government, or was it part of a larger pattern of Justices who depart from executive preferences no matter who controls the senate?
Unfortunately, the data available to previous researchers have been too limited to answer this question. Leading Supreme Court scholars have been "especially handicapped" in their ability "to offer information on voting behavior prior to the Vinson Court era." (12) Indeed the "greatest single resource of data on the Court,"13 Harold J. Spaeth's U.S. Supreme Court Database, offers Justice-centered voting records dating back to only 1946.14 With just twenty-seven Supreme Court appointments since this time, the sample of Justices whose voting records are available to empirical legal scholars is too small to draw statistically confident conclusions about the president's probability of failure. (15) Scholars have begun making inroads by looking as far back as the 1930s. (16) Still, existing studies leave almost twice as many appointments unexamined as examined. They omit several appointments made during earlier periods of divided government.
Our Article addresses this shortcoming. We introduce a new data set containing the Supreme Court voting alignments for every written opinion issued from 1838 through 1949. This data set contains 18,812 cases and represents an approximate three-fold increase in quantitative voting data available to Supreme Court scholars. We build on past historical studies17 and make possible the first comprehensive study of precise voting coalitions for early Supreme Court cases. (18)
Our new data allow us to address previously unanswerable questions about the president's power to appoint Justices with the Senate's "Advice and Consent." (19) A powerful Senate could either keep a nominee off the Court or drive the president to nominate a more moderate Justice than he would otherwise prefer. Historical accounts of failed nominees make clear that we cover a contentious period in which divided government often led to a nominee's defeat. (20) We know far less, however, about the Justices who made it on the Court. Did the confirmation process make it more difficult for a president to appoint Justices who shared his ideology?
The literature on Supreme Court appointees is rife with competing anecdotes of executive success rates. To be clear, our discussion of satisfaction or disappointment in overall voting records focuses on a single measure, ideology. It does not account for merit or other factors relevant to evaluating a Justice. But assessments of ideology alone provoke considerable differences of opinion. Some scholars highlight presidents disappointed with their appointees' treatment of key executive policies, (21) and luminaries such as John Hart Ely have wondered whether disappointment should be considered the "rule rather [than] the exception." (22) Others claim that these accounts are "more myth than reality" (23) or find it difficult to set a clear benchmark for measuring executive disappointment.24 Still others blame disappointment on the constraining effect of an opposing-party Senate25 or warn against omitting the Senate when crafting proxies for an appointee's ideology. (26)
Our extended time period allows us to compute clear benchmarks and conduct the first quantitative examination of presidential success rates over a 170-year period. (27) We proceed as follows: In Part II, we describe our data collection process and methodology for measuring voting alignments. We also show that our measures of agreement based on count data reflect the information contained in another leading measure of judicial ideology, Martin-Quinn scores.
In Part III, we present our general results. Aggregate voting records show that Justices generally agree with appointees of the same party about as often as they agree with appointees of the other party. Most surprising, however, are the voting records of individual Justices. They show that presidents fare far better with certain appointees than others, and that they often have appointed ideologically incompatible Justices. Just under half of the Justices we studied sided with appointees of the other party most of the time.
We find no evidence that this outcome can be attributed solely to divided government, and most of our results remain stable over time. In very recent years, though, the Court has become more polarized as the magnitude of Justices' agreement with one side or the other has increased. Thus, while the stakes in a given appointment are higher than ever, history gives no assurance that a Justice will side with the party of his or her appointing president.
In Part IV, we apply these individual voting records to three lines of historical inquiry. First, we assess prominent examples of disappointment, including Justice Salmon Chase in 1870 and Justice Oliver Wendell Holmes in 1904. As Felix Frankfurter recounted, these Justices let down Presidents Abraham Lincoln and Theodore Roosevelt on key executive policies:
Chief Justice Chase declared unconstitutional the Legal Tender Act of the President who appointed him and of the Administration of which he was a member. So also Mr. Justice Holmes decided against the Government in the Northern Securities case, Roosevelt's pet litigation, although the latter thought it his duty to put on the Supreme Bench only men who would sustain "My Policies." (28)
Historical accounts differ as to whether Holmes let down Roosevelt in just the Northern Securities case or with his voting record as a whole. (29) We also do not know how his complete voting record compares to that of Justice William Day, who also was thought to have disappointed Roosevelt. overall voting records allow us to identify whether these and other executive disappointments hinged on a few key cases or on overall voting records.
Our second historical inquiry is a comprehensive examination of individual voting records for Justices nominated during times of divided government in the nineteenth century. Again, our data offer the first opportunity to assess Justices' overall voting records during this important but previously inaccessible time period. Like our general findings, these records fail to show that divided government led to Justices with significantly more moderate voting records.
Our last historical inquiry is a preliminary measure of whether past voting patterns are consistent with ideological drift. This phenomenon, in which Justices' ideologies change over time, has been documented in contemporary data. (30) Our records show that Justices' historical voting patterns also are consistent with ideological drift.
In Part V, we conclude that a large percentage of Justices have disappointed presidents with their overall voting records. This outcome has continued to occur at a high and stable frequency since 1838. We find no correlation between an opposing-party Senate and the rate of appointees who vote across party lines. Indeed, many prominent historical examples of failed appointments occurred when the Senate and president were of the same party. These mavericks are not outliers. Instead, they are part of a larger pattern of appointees whose votes depart or drift away from executive expectations more often than voters may think.
We compiled a comprehensive record of votes for every written opinion the Supreme Court issued from 1838 to 1949. There are 18,812 case citations for this time period. We relied on the previous work of the Supreme Court Historical Society and eighteen University of Michigan students to help gather and code voting records.
We tailored our data collection strategy to meet two primary goals. The first goal was to facilitate comparison with leading studies of contemporary voting alignments. The second goal was to ensure accuracy by eliminating as much discretion as possible in our students' coding. (31) As a further guarantee of accuracy, we used a double-entry system that required two students to independently review and code each entry.
With these goals in mind, we assembled our database in four main steps:
1. To obtain a complete list of written opinions that the Court issued between 1838 and 1949, we consulted the Case Citation Finder on the United States Supreme Court's website. (32) It sets forth the official citations, "in the form recommended by the Reporter of Decisions, for every signed, per curiam, or in-chambers opinion published (or soon to be published) in the United States Reports."33 This listing yielded 18,812 case citations for our review.
2. Next, we identified the subset of non-unanimous cases. To do so, we used an index created by the Supreme Court Historical Society to identify every case in which any Justice authored a non majority opinion. (34) The index provides a comprehensive list of opinions authored by each Justice. (35) It further classifies the opinions into seven different categories: opinions of the Court, opinions announcing judgment, concurrences, dissents, separate opinions, statements, and opinions authored as circuit justices. (36)
3. The index identified 2,117 non-unanimous cases that the Supreme Court's Case Citation Finder listed for our time period. For each of these non-unanimous cases, we turned to LEXIS's on-line database (37) to identify specific Justices joining each concurrence, dissent, statement, or separate opinion. To ensure accuracy we hired eighteen student research assistants to double enter the voting coalitions. We assigned each case to two students via a random assignment algorithm. The students did not know which other student received the same case.
We instructed the students to read each case in LEXIS and code the votes in machine-readable form. We wrote code in MATLAB (38) to compare coding across students and flag cases in which the coding disagreed. Cases with disagreement were forwarded to a third student. If this student's coding agreed with one of the original two, the coding was accepted. If the votes remained unclear, one of the authors of this Article read the case and coded votes accordingly.
4. For all opinions announcing judgment or majority opinions (including those in unanimous cases), we coded voting coalitions by listing all Justices currently on the Court who had not joined or authored another opinion.
We drew information about the president and the Senate at the time of nomination from the U.S. Supreme Court Justices Database.39 This database documents (1) the name of the nominating president, (2) the political party affiliation of the nominating president, and (3) the dominant political party of the U.S. Senate at the time of nomination. (40) For all Justices who were nominated to Chief Justice at the time they were sitting as an Associate Justice, we use data reflecting the Justice's initial appointment to the Court. (41)
Readers interested in a detailed description of our data collection process should consult the Appendix.
A. Our Data Set Compares Directly to a Well-Defined Subset of the U.S. Supreme Court Database
The 18,812 decisions we code represent a substantial body of the Court's work-and also one most likely to reveal philosophical divides between Justices. our data compare directly to a clearly-defined subset of cases in the U.S. Supreme Court Database. Like our data set, the Supreme Court Database reports voting records arranged by case citation. (42) The Supreme Court Database covers all "full opinion cases" as well as all per curiam opinions for cases in which the Court held an oral argument or provided a summary or opinion explaining its reasoning. (43)
These similarities allow us to link our historical measure of the president's appointment power to contemporary voting records in the U.S. Supreme Court database. (44)
B. Our Measure of Agreement Based on Count Data Captures Much of the Information Reflected in Martin-Quinn Scores
our measure of voting behavior is straightforward and time-honored. (45) We count (a) the total number of times Justices agreed and (b) the total number of opportunities these Justices had to agree. Here, Justices agree when they both vote for the majority or minority on a judgment. Thus, we employ a simple formula to calculate percentage of agreements: percentage of agreement = (a) number of times Justices agreed/(b) number of opportunities to agree. This percentage allows us to see who votes together and at what rate. For purposes of our inquiry, we also can compare the percentage of times a Justice agrees with appointees of the same president or party to percentages of agreement with appointees of a different president or party.
To measure percentages of agreement between the Justices, we consider each opportunity two Justices had to agree in a particular case. In a case decided by nine Justices, there are thirty-six total opportunities to agree. To illustrate this point, Figure 1 below shows how votes are counted for the landmark First Amendment case, Abrams v. United States. (46) The decision split 7-2 with Justices Clarke, White, Pitney, Day, Van Devanter, McReynolds, and McKenna voting to affirm and Justices Holmes and Brandeis dissenting. (47)
Here Justice Clarke agreed with his own party two out of three opportunities (67%) and with the other party four out of five opportunities (80%). By counting agreements between different pairs of Justices, we can calculate the percentage of times they vote with different groups of appointees in all cases.
The percentage of agreement for the entire group of non-unanimous cases shows whether appointees of certain presidents or parties are more likeminded than others. of course, this percentage alone does not demonstrate presidential success, as a president could be disappointed in the similar voting records of all his appointees. This concern vanishes when voting records are viewed in context, as we do in Parts Ill and IV, below. The literature rarely asserts that a president has failed to appoint ideologically compatible Justices in every case. And failure never extends to Justices of an entire party. (48)
Our voting analysis captures much of the information reflected in a leading measure of judicial ideology, Martin-Quinn scores. (49) Andrew Martin and Kevin Quinn use observable agreement data-votes for a majority or minority on the judgment (50)-to fit a parametric model. (51) The model maps Justices' voting behavior to a set of numbers, or "ideal points," that can be arranged on the real line from closest to farthest apart. (52) Justices who vote together frequently (such as Scalia and Thomas) are close together on the line. Justices who seldom vote together (such as Breyer and Thomas) are farther apart. As Ward Farnsworth explains:
The [Martin-Quinn] model studies all the …